Moonlighting over the business cycle.

Amuedo-Dorantes, Catalina ; Kimmel, Jean

I. INTRODUCTION

During economic downturns, employment levels fall, unemployment rates increase, and real wages drop. As such, real wages are considered somewhat procyclical. Yet, very little is known about the cyclicality of multiple-job holding. From a graphical analysis of national time-series moonlighting data during the 1960s and 1970s, Stinson (1987) finds evidence of large increases in moonlighting during expansionary periods. Likewise, Partridge's (2002) findings support the possibility of procyclical multiple-job holding. Yet, the popular media and employment think tanks discuss moonlighting as a by-product of financial pressures. (1) Furthermore, the idea that individuals may hold multiple jobs during an economic downturn in order to supplement family income is consistent with the so-called "added worker effect" in the labor economics literature.

Why should we care about the cyclicality of multiple-job holding in the economy? There are numerous reasons. First, moonlighting has played, and can be expected to continue to play, a visible role in the U.S. workforce. The moonlighting rate was 5.2% as of 1970, with 7.0% of male workers and 2.2% of female workers holding multiple jobs. This moonlighting rate remained practically unchanged over the course of the next 30 yr. Yet, its gender incidence did fluctuate. In particular, the female moonlighting rate grew over the past several decades from the above listed 2.2% in 1970, to 3.8% in 1980 and to 5.9% in 1991, and exceeded the male rate for the first time in 1995 (6.5% vs. 6.3%). In May 2007, the moonlighting rate was still similar to the moonlighting rate in 1970 (i.e., 5.3%), although now females are much more likely to moonlight than men, with rates of 4.9% vs. 5.7%. (2) Considered over the course of an entire year (vs. at a point in time), an even larger percent age of workers hold second jobs. In this regard, Paxson and Sicherman (1996, p. 357) note that approximately 20% of male workers and 12.2% of female workers hold second jobs in any given year. (3) As these numbers suggest, moonlighting is a substantively important labor market phenomenon; however, it is much understudied. In fact, our review of the literature to follow will show that only a handful of research papers have studied this phenomenon. Thus, our primary motivation for writing the paper is to shed light on a poorly understood, yet numerically important, feature of the labor market.

A second reason for examining the cyclicality of multiple-job holding is the potential importance of moonlighting in facilitating labor supply adjustments during temporary economic downturns or upturns. (4) Through a description of moonlighters, our analysis will shed light on how workers respond to fluctuations in job opportunities across the business cycle. Macroeconomists and labor micro-economists have long studied employment cyclicality in an attempt to improve the explanations of the changing composition of the labor force over business cycles, for example, the added and discouraged worker effects. (5)

Third, as one part of the broader picture of economic cyclicality, our study fits nicely into the ongoing debate concerning the cyclicality of real wages. (6) Considerable research effort has been devoted to understanding the nature of real-wage fluctuations across business cycles, yet these studies are muddied by a glossing lack of distinction between workers' average real wages across all jobs and their wages in the primary job alone. Understanding the role that moonlighting may play in employment cyclicality may inform the debate concerning the measurement of real wages when studying real-wage cyclicality.

A final and policy-relevant motivation for this research refers to the structure of employment taxes affecting secondary jobs. As explained by Anderson and Meyer (2003), unemployment insurance (UI) payroll taxes are highly regressive, in large part because benefits are structured in the same way. (7) However, due to the UI payroll tax's low taxable wage base, moonlighters working few hours in the second job are still subject to the full UI tax despite likely lacking eligibility for benefits in the event of a layoff. Considering the important role that multiple-job holding can play in the economy by responding to firms' "just-in-time" labor needs, a better understanding of the cyclical nature of moonlighting can help inform the debate on how to best structure the UI tax. (8)

Given the magnitude of moonlighting and the policy implications that its cyclicality may have for the functioning of the labor market, we examine the responsiveness of male and female multiple-job holding to business cycles. We first review the existing descriptive evidence regarding this question. Subsequently, we describe our data and methodology, concluding with a discussion of our findings.

II. BACKGROUND ON THE CYCLICALITY OF MOONLIGHTING

There is often the presumption that moonlighting is countercyclical. In this vein, Mishel, Bernstein, and Schmitt (1999) of the Employment Policy Institute asserted that "The benefits of persistent low unemployment are reflected in many labor market indicators. Multiple job holding, for instance, has fallen over the last year...." Nonetheless, from a theoretical standpoint, moonlighting rates can be procyclical or countercyclical. Focusing on economic downturns, from a demand side, moonlighting opportunities may be limited during a recession as the total number of jobs falls. On the other hand, from a supply side, workers may choose to moonlight in an effort to stabilize family income during an economic downturn when unemployment rates are higher and real wages lower.

Partridge (2002) examines moonlighting during the periods 1994 and 1998 using state-level data. While his focus is on the nature of second-job holding during a period of strong economic growth, his paper offers insight into the potential cyclical pattern of moonlighting. If moonlighters face a relatively high likelihood of being laid off during periods of high unemployment, or if moonlighting rises during periods of rapid economic growth and labor shortages, then moonlighting might be procyclical (p. 426).

Conway and Kimmel (1998) propose a model that leads to a more rigorous prediction regarding the cyclicality of moonlighting. According to their theoretical framework, hours on the primary job and hours on the secondary job (along with leisure hours) enter the utility function separately. Their model explicitly allows for two distinct reasons for moonlighting: primary-job constraints (e.g., underemployment) and job heterogeneity (i.e., the second job might provide nonwage remuneration or affect utility differentially from the primary job). From their model, it follows that an increase in nonwage income leads to a decline in moonlighting. As such, moonlighting might be countercyclical. Via the estimation of a fixed-effects logit model of the likelihood of holding multiple jobs, Heineck and Schwarze (2004) find support for this notion. (9) However, Conway and Kimmel (1998) estimate a positive second-job labor supply elasticity for men, suggesting that moonlighting might move cyclically with the business cycle since wages have been found to be slightly procyclical. (10)

Renna's (2006) cross-country comparative research on moonlighting and overtime work provides interesting insight into our question regarding the cyclicality of moonlighting. Renna finds that the incidence of moonlighting increases in the face of increased primary-job constraints imposed by declining standard hours of work. In other words, as workers have found it more difficult to accommodate desires for extra work hours in their primary jobs, they are more likely to take second jobs. This implies that for workers seeking to increase their total work hours, moonlighting could serve as an alternative to overtime work. Thus, his finding that overtime work is procyclical suggests that moonlighting may also be procyclical as both employment options are responses to the similar desire for increased work hours.

Two other possible motivations for moonlighting may be individuals' responses to negative financial shocks and to primary-job insecurity. Boheim and Taylor (2004) examine these moonlighting reasons, both closely related to business cycle fluctuations, and find mixed evidence of multiple-job holding in response to financial shocks and weak support for the job insecurity motivation.

Yet, other studies in the literature refer to expectations about future income as another motive for moonlighting. In this vein, Bell, Hart, and Wright (1997a, 1997b) investigate the possibility that workers might take second jobs as a hedge against future unemployment. That is, as expectations regarding a future economic downturn rise, moonlighting rates might increase. However, their study, which uses British data from 1991 to 1998, fails to yield support for this hypothesis. (11)

In sum, there are sufficient reasons and evidence to suggest that moonlighting rates may respond in some systematic fashion to cyclical labor market fluctuations, but there is no clear indication of whether multiple-job holding is pro- or countercyclical. As such, we also lack a definite prediction regarding gender differences in moonlighting cyclicality. Given that men and women exhibit different labor market trajectories and different degrees of employment cyclicality, their moonlighting choices and cyclicality may vary. (12) Indeed, historically, women were more likely to moonlight for financial reasons as described by Kimmel and Powell (1999) due to primary-job constraints as described by Averett (2001) or to meet family responsibilities as described by Allen (1998). These gender differences in moonlighting may have diminished over time as female moonlighting rates have come to mirror those of men (and exceed male rates in recent years). Yet, moonlighting women are still much more likely to work in a full-time primary job and a part-time secondary job, while moonlighting men are more likely to hold two full-time jobs.

Why should we expect the cyclicality of moonlighting to differ between men and women? One could cite various reasons. First, gender differences in moonlighting cyclicality could be due to differences in the demographics of male and female moonlighters. For instance, male moonlighters are more likely to be married or have children, while the opposite is true for women. Second, gender differences in moonlighting cyclicality could stem from occupational segregation by sex and/or from industry seasonality. In this vein, Goodman (2001) examines the cyclicality of service jobs and notes that service jobs do not suffer much during recessions (although their growth does wane). To the extent that women are more likely to work in the service sector, they may exhibit less cyclicality in their moonlighting behavior than men. However, Hotchkiss and Robertson (2006) show that lesser educated workers, many of whom are employed in the service industry, are more responsive to business cycle fluctuations. Finally, Stinson (1990) shows that men moonlight for long periods of time and women are more likely to moonlight on a temporary basis to meet financial needs. This feature may have some implications for gender differences in moonlighting cyclicality as well.

Much of the above-described gender differences in moonlighting, because they are explained by observable demographic differences, will be eliminated when standard regression procedures are implemented. Thus, the question arises that, once observable factors are controlled in a regression framework, would we expect any remaining gender differences in moonlighting? First, to the extent that we cannot measure precisely an individual's reservation wage, we might expect such gender differences to persist. But, what if regression methodologies are used that adjust, in some sense, for unobservable differences across individuals? Then, we might expect some portion of expected gender differences in moonlighting patterns to disappear.

III. DATA AND DESCRIPTIVE STATISTICS

We use data drawn from the Geo-coded 1979 National Longitudinal Survey of Youth File (NLSY79 Geo-coded file). (13) This is a nationally representative sample of 12,686 civilian young men and women aged 14-21 yr as of December 31, 1978. This cohort was initially interviewed annually from 1979 through 1994. Starting in 1994, the interviews were conducted biennially through the year 2002.

We have chosen the NLSY79 survey because its questionnaire is best suited for analyzing our research question. (14) Specifically, the survey instrument permits individuals to report up to six distinct jobs held within each calendar year and includes job details, such as job start and end dates. This permits us to construct precise measures of multiple-job holding. Other data sets, such as the Panel Study of Income Dynamics, may result in an overstatement of moonlighting, as it is difficult to distinguish multiple-job holders from job changers. The one drawback with the NLSY79 is that the data are representative of a specific-age cohort. However, focusing on this prime-aged sub-sample allows us to derive useful information regarding moonlighting cyclicality and partially alleviates heterogeneity concerns.

We work with separate unbalanced panels of men and women from the 20 rounds of the NLSY79. (15) In 2002, a total of 8,033 civilian and military respondents were interviewed. We restrict our sample to person-year observations for which information is available regarding employment, earnings, race, gender, age, education, marital status, fertility, work limitations due to health-related reasons, and other location-specific variables. We use the week-by-week longitudinal work records on each respondent from January 1978 to the year 2002 to construct variables indicative of the respondent's sector of employment, occupation, tenure, weekly hours worked, and hourly rate of pay at the primary job. (16) Similarly, we create a dummy variable indicative of whether the respondent moonlighted, which we define as holding more than one job simultaneously for longer than 1 wk. (17)

Preliminary employment and moonlighting rates for men and women over the past two decades are shown in Table 1. We compare employment and moonlighting rates for three time periods centered around 1980, 1990, and 2000. (18) In part due to the aging nature of the NLSY79 cohort, employment and moonlighting rates increased between the 1980 and the 1990 time period and then declined between the two time periods centered around 1990 and the year 2000. In particular, while moonlighting rates averaged 7% for both men and women over the period under consideration, moonlighting rates showed some fluctuations, peaking around 1990 for both men and women. Note that this now prime-aged sample exhibits a moonlighting rate somewhat above the national average of 5.3% for all workers.

Table 2 Panels A and B inform on some of the personal characteristics of single- and multiple-job holders for the same periods displayed in Table 1. An increasingly larger fraction of multiple-job holders are black, with the percentage of moonlighters in our NLSY79 sample who are black rising from 14% in the time period centered around 1980 to 32% around 2000. Additionally, moonlighters appear to be more highly educated than non-moonlighters, although the education gap narrows over time. Looking at family characteristics, married men and women are less likely to hold more than one job, although the difference is quite small, and male single-job holders seem to have more children than their moonlighting counterparts in earlier decades, Finally, a higher fraction of moonlighters reside in urban areas relative to single-job holders.

IV. METHODOLOGY

Our purpose is to examine the cyclicality of male and female moonlighting during the 1980s and 1990s up to the year 2002. Underlying our empirical analyses is a standard individual utility-maximizing model, (19) according to which working men and women decide whether to moonlight and, if so, the number of hours they will work in more than one job. Note, however, that a nonnegligible number of working men and women do not moonlight. Therefore, the distribution that applies to the sample data is a mixture of discrete and continuous distributions, rendering the use of ordinary least squares (OLS) inappropriate. Consequently, a Tobit model would seem more appropriate as it would take into account the censored nature of the distribution of working men and women's moonlighting hours by modeling the likelihood of moonlighting and the hours moonlighted as a function of the same covariates.

A potential disadvantage of the Tobit model is that a change in any regressor will have the same overall effect (i.e., the same sign) on both the probability of moonlighting and the number of hours moonlighted. Hence, a two-part model could improve on the estimation by allowing for the possibility that variables affecting the decision to moonlight may impact the hours moonlighted differently. Nonetheless, recognizing (a) the difficulty of conceiving appropriate identifiers that affect the decision to moonlight without influencing the hours moonlighted and (b) the sensitivity of the findings to the choice of identifiers inherent in the estimation of two-part selection models, we view the estimation via a Tobit model as preferable. (20)

As such, we estimate the following random-effects Tobit model. (21) The random-effects specification allows us to adjust standard errors for the group-wise heteroskedasticity arising from the fact that the growth rate of nonfarm employment only varies across states, while the remaining variables in our model vary across individuals, while also taking into account some of the individual heterogeneity shaping male and female moonlighting rates. We follow the correction methodology outlined by Moulton (1986). Thus, we model the likelihood and the number of hours moonlighted by men and women as follows:

[y.sup.*.sub.it] = [[alpha].sub.i] + [X.sub.it] [beta] + [[epsilon].sub.it],

where [y.sub.it] = max (0, [y.sup.*.sub.it])

and [[epsilon].sub.it] | [X.sub.it], [[alpha].sub.i] ~ N (0, [[sigma].sup.2.sub.e]).

The vector [X.sub.it] controls for a variety of personal, family, regional, and time-related factors and specific primary-job characteristics known to affect male and female employment patterns. In particular, among the personal and family characteristics, we include two dummies for race (black and other race), a continuous measure of age, the highest grade completed by the respondent, a dummy variable for marital status (married), and two measures of parental status (a dummy variable for the presence of young children in the household plus a continuous count of the total number of children in the household). One might expect that individuals with greater demands on their nonmarket time (i.e., higher reservations wages) might, in addition to the standard conclusion of being less likely to seek paid employment, also be less likely to moonlight. As we described earlier, we expect that these demographic controls will eliminate some of the observed gender differences in moonlighting.

We also include a variety of primary-job characteristics possibly affecting the moonlighting decision and the hours moonlighted, such as the real primary-job hourly wage, tenure, occupation, and whether the job is in the private or public sector. Likewise, we incorporate a number of regional and time-related factors in our regression analysis. For instance, to control the fact that wealthier states might exhibit systematically different moonlighting patterns than less wealthy states, we include the state's per capita income. Additionally, we include a dummy variable for residing in an urban area and three regional dummies to address other macroeconomic differences in job markets. Regarding time, we include a continuous time trend measure plus three time period dummies for the periods of time: 1979-1985, 1986-1992, and 1993-1999, with the period 2000-2004 as the excluded period. (22) We structure our four time periods in this way so as to reflect distinct economic circumstances. The first and second periods wrap around a recession and, thus, include the buildup to the recession, the recession itself, and the recovery. The third period is unique as it reflects a 7-yr period of substantial economic growth. Finally, the fourth period is the shortest of the four and captures a slight economic downturn following the rapid economic growth of the mid- to late 1990s. We would expect period differences in moonlighting; therefore, structuring our periods in this way will maximize our efforts to reveal these differences.

Finally, our business cycle measure is the state's growth rate in nonfarm employment. The bulk of the literature examining the cyclicality of real wages relies on national unemployment rate measures, insufficient for our needs due to the existence of regional disparities in industrial composition and moonlighting rates (see, e.g., Partridge 2002). Additionally, state unemployment rates may be subject to greater measurement error due to the smaller sample sizes from which the data are drawn. Finally, as described by Blanchard and Katz (1992), reliance on a static measure of economic activity at the state level may be problematic because states vary in their equilibrium unemployment rates. Thus, we rely on the more accurately measured nonfarm employment and construct its growth rate. Additionally, to better capture potential differences in moonlighting cyclicality over the time period under examination, we interact the time period dummies with the growth rate in nonfarm employment.

V. MOONLIGHTING OVER THE BUSINESS CYCLE

Results from our random-effects Tobit models are displayed in Table 3 Panels A and B. Male and female blacks moonlight more than whites. Married men, men with greater family responsibilities, and men and women residing in states with higher per capita incomes moonlight less than single men, men with fewer children, or men and women residing in poorer states, respectively. In contrast, more educated men and women seem more likely to moonlight, perhaps due in part to their access to part-time consulting opportunities.

Primary-job characteristics play an important role in the decision to moonlight, particularly among women. For instance, both men and women appear more likely to hold a second job, the longer their tenures in their primary jobs. Perhaps workers with longer job tenures are more reluctant to change jobs entirely when new opportunities arise, resulting in higher multiple-job holding. Additionally, men and women working in higher skilled occupations seem less likely to moonlight than their counterparts in service-related jobs (the category of reference). It is also worth noting that, while male moonlighting does not seem to be shaped by primary-job wages, female moonlighting is responsive to wage changes. A $10 increase in women's primary-job wages would reduce their moonlighting likelihood by 1 percentage point and hours moonlighted by approximately 0.2 h. This is consistent with the finding of Conway and Kimmel (1998). In addition to wages, women working in the public sector are about 6 percentage points more likely to moonlight and, once they moonlight, work two more hours in their secondary jobs than their counterparts in the private sector.

Of particular interest are the differences in moonlighting over the various periods of time. Male and female moonlighting were more prominent during each of the three time periods spanning between 1979 and 1999 relative to the 2000-2002 period used as reference. Specifically, during the 1979-1985 period, men and women were about 12 and 21 percentage points more likely to moonlight, respectively. (23) If they held multiple jobs, moonlighting men and women worked an average of four and eight more hours per week than their counterparts during the more recent reference period of 2000-2002. These estimates dropped between 1986 and 1992 to 10 percentage points and 3 h among men and to 15 percentage points and 5 h among women. Finally, between 1993 and 1999, men and women were only 7 and 6 percentage points, respectively, more likely to moonlight than their counterparts in 2000-2002. Additionally, their hours moonlighted dropped to just two more hours per week.

What can we say about the cyclicality of moonlighting? Period-specific moonlighting cyclicality results for men and women are presented in Table 4. There is no evidence of any moonlighting cyclicality for men. However, female moonlighting is countercyclical during the 1980s and early 1990s and then turns procyclical by the beginning of the twentieth century. Specifically, a 2% increase in the growth rate of nonfarm employment (the average in our sample is close to 2%--see Appendix Table A1) would lower the probability of moonlighting among women by approximately 0.8 percentage points in 1979 1985 and by 1.4 percentage points during 1986-1992. However, the same increase in the growth rate of nonfarm employment would increase the likelihood of moonlighting and the weekly hours moonlighted by women by approximately 3 percentage points and 58 min, respectively, during the reference time period of 2000-2002. These results help explain the apparently contradictory views of moonlighting as a by-product of economic distress stressed by advocacy groups with the view of moonlighting as the response of just-in-time labor to an increasing labor demand following periods of economic growth.

VI. SUMMARY AND CONCLUSIONS

We examine male and female moonlighting cyclicality over the 1980s, 1990s, and early twentieth century. We find that, despite diminishing over time, both men and women were more likely to moonlight and moonlighted longer hours during the 1980s and 1990s than during the 2000-2002 period. Additionally, while male moonlighting does not seem to respond to business cycles, female moonlighting does. Specifically, consistent with the popular media, female moonlighting appeared countercyclical during the 1980s and early 1990s. These are time periods that wrapped up around a recession, and as such, our finding suggests that moonlighting during these times may have been a by-product of economic distress. This finding is also consistent with the lower incidence of moonlighting in higher per capita income states. Yet, this countercyclical behavior disappears during the 19931999 perio--a period of rapid economic growth to become procyclical by the early twentieth century. The recent procyclicality of female moonlighting, which follows the economics growth of the 1993-1999 period, supports the idea that female workers respond to a need for just-in-time employment characteristic of economic upturns. While the analyses differ in their focus and usage of state-level data, the recent procyclicality of female moon lighting supports the findings by Partridge (2002) and Renna (2006). Partridge (2002) argues that short-run moonlighting appears to be procyclical and states that "moonlighting appears to be a regional labor market shock absorber" (p. 438). Similarly, Renna (2006) finds that, like overtime work, moonlighting is procyclical and is used by workers as a means to increase their work hours.

Overall, the analysis provides us with a better understanding of the variability and business cyclicality of male and female moonlighting over the past decades, which can prove useful in anticipating the adjustment of men and women to fluctuations in job opportunities over the business cycle. Additionally, our findings are relevant for the literature on real-wage cyclicality. To the extent that the incidence of moonlighting is procyclical during recent years and primary-job wages exceed secondary-job wages, (24) average hourly real wages across all jobs should be more procyclical than primary-job wages due to the incidence of secondary jobs. Further research examining real-wage cyclicality separately by gender across all jobs held concurrently may help assess the role of moonlighting cyclicality in explaining overall real-wage cyclicality. Finally, a better understanding of the cyclicality of male and female moonlighting can prove useful in informing the debate on how to structure UI taxes.

ABBREVIATIONS

NLSY79: 1979 National Longitudinal Survey of Youth OLS: Ordinary Least Squares

UI: Unemployment Insurance

doi: 10.111l/j.1465-7295.2008.00140.x

APPENDIX TABLE A1
Means and Standard Deviations (SD)

 Women

 One Job Moonlighting

Variables Mean SD

White 0.69 0.46
Black 0.25 0.43
Other race 0.05 0.22
Age 27.49 6.58
Highest grade 12.80 2.18
Married 0.46 0.50
Separated 0.05 0.21
Divorced 0.09 0.29
Widowed 0.00 0.07
Never married 0.40 0.49
Young children 0.32 0.47
No. of children 0.94 1.12
Health limitations 0.05 0.22
Previous year's nonlabor income 17,190.75 40,122.81
Urban 0.80 0.41
Northeast 0.17 0.38
North central 0.23 0.42
South 0.40 0.49
West 0.19 0.39
Time period 1979-1985 0.38 0.48
Time period 1986-1992 0.37 0.48
Time period 1993-1999 0.18 0.38
Time period 2000-2002 0.08 0.27
State per capita income 17,759.55 6,326.97
Growth nonfarm employment 1.90 2.02
Real hourly wage 5.98 10.88
Public sector job 0.14 0.34
Professional, technical, clerical 0.54 0.50
Craftsmen, operatives, laborers 0.13 0.34
Sales and services 0.30 0.46
Tenure 33.38 34.74
Hours in nonprimary job -- --

 Women

 One Job Moonlighting

Variables Mean SD

White 0.71 0.45
Black 0.24 0.43
Other race 0.05 0.21
Age 28.42 6.46
Highest grade 13.02 2.33
Married 0.41 0.49
Separated 0.03 0.17
Divorced 0.07 0.26
Widowed 0.00 0.03
Never married 0.48 0.50
Young children 0.24 0.43
No. of children 0.67 1.07
Health limitations 0.03 0.18
Previous year's nonlabor income 13,476.70 40,006.20
Urban 0.80 0.41
Northeast 0.19 0.39
North central 0.27 0.45
South 0.33 0.47
West 0.20 0.40
Time period 1979-1985 0.30 0.46
Time period 1986-1992 0.39 0.49
Time period 1993-1999 0.22 0.41
Time period 2000-2002 0.10 0.30
State per capita income 18,900.42 6,209.25
Growth nonfarm employment 1.89 1.95
Real hourly wage 7.92 13.72
Public sector job 0.15 0.36
Professional, technical, clerical 0.35 0.48
Craftsmen, operatives, laborers 0.40 0.49
Sales and services 0.22 0.41
Tenure 56.57 83.19
Hours in nonprimary job 30.49 25.32

 Men

 One Job Moonlighting

Variables Mean SD

White 0.69 0.46
Black 0.25 0.44
Other race 0.05 0.23
Age 27.35 6.54
Highest grade 12.41 2.37
Married 0.40 0.49
Separated 0.03 0.18
Divorced 0.06 0.24
Widowed 0.00 0.04
Never married 0.51 0.50
Young children 0.24 0.43
No. of children 0.63 1.04
Health limitations 0.03 0.18
Previous year's nonlabor income 13,009.20 37,993.54
Urban 0.79 0.41
Northeast 0.18 0.38
North central 0.24 0.42
South 0.38 0.48
West 0.20 0.40
Time period 1979-1985 0.38 0.49
Time period 1986-1992 0.36 0.48
Time period 1993-1999 0.18 0.38
Time period 2000-2002 0.08 0.27
State per capita income 17,795.71 6,300.73
Growth nonfarm employment 1.90 2.04
Real hourly wage 11.57 280.72
Public sector job 0.08 0.27
Professional, technical, clerical 0.25 0.43
Craftsmen, operatives, laborers 0.53 0.50
Sales and services 0.19 0.40
Tenure 33.93 62.79
Hours in nonprimary job -- --

 Men

 One Job Moonlighting

Variables Mean SD

White 0.72 0.45
Black 0.23 0.42
Other race 0.04 0.20
Age 28.63 6.73
Highest grade 13.52 2.18
Married 0.39 0.49
Separated 0.04 0.20
Divorced 0.11 0.32
Widowed 0.01 0.08
Never married 0.45 0.50
Young children 0.22 0.41
No. of children 0.84 1.14
Health limitations 0.05 0.22
Previous year's nonlabor income 16,995.04 38,280.87
Urban 0.81 0.40
Northeast 0.20 0.40
North central 0.26 0.44
South 0.36 0.48
West 0.18 0.38
Time period 1979-1985 0.31 0.46
Time period 1986-1992 0.37 0.48
Time period 1993-1999 0.22 0.41
Time period 2000-2002 0.11 0.31
State per capita income 18,902.96 6,395.50
Growth nonfarm employment 1.86 1.94
Real hourly wage 6.82 13.94
Public sector job 0.18 0.38
Professional, technical, clerical 0.60 0.49
Craftsmen, operatives, laborers 0.11 0.31
Sales and services 0.26 0.44
Tenure 55.27 65.89
Hours in nonprimary job 25.67 22.08

REFERENCES

Allen, W. D. "The Moonlighting Decision of Unmarried Men and Women: Family and Labor Market Influences." Atlantic Economic Journal, 26, 1998, 190-205.

Anderson, P. M., and B. D. Meyer. "Unemployment Insurance Tax Burdens and Benefits: Funding Family Leave and Reforming the Payroll Tax." National Bureau of Economic Research Working Paper No. 10043, 2003.

Averett, S. L. "Moonlighting: Multiple Motives and Gender Differences." Applied Economics, 33, 2001, 1391-1410. Bell, D. N. F., R. A. Hart, and R. E. Wright. "Double Job Holding in Great Britain." Unpublished manuscript, 1997a.

--. "Multiple-Job Holding as a 'Hedge' Against Unemployment." CEPR research programme on UK Labour Market: Microeconomics Imperfections and Institutional Features, 1997b.

Blanchard, O. J., and L. F. Katz. "Regional Evolutions." Brookings Papers on Economic Activity, 1, 1992, 1-75.

Boheim, R., and M. Taylor. "And in the Evening She's a Singer with Band--Second Jobs, Plight or Pleasure?" IZA Discussion Paper No. 1081, 2004.

Conway, K., and J. Kimmel. "Male Labor Supply Estimates and the Decision to Moonlight." Labour Economics, 5, 1998, 135-66.

Devereux, P. J. "The Cyclicality of Real Wages within Employer-Employee Matches." Industrial and Labor Relations Review, 54, 2001, 835-50.

Goodman, W. C. "Employment in Service Industries Affected by Recessions and Expansions." Monthly Labor Review, 124, 2001, 3-11.

Hart, R. A. "Real Wage Cyclicality of Female Stayers and Movers in Part-Time and Full-Time Jobs." IZA Discussion Paper No. 2364, 2006.

Heineck, G., and J. Schwarze. "Fly Me to the Moon: The Determinants of Secondary Job Holding in Germany and the U.K." IZA Discussion Paper No. 1358, 2004.

Hotchkiss, J. L., and J. C. Robertson. "Asymmetric Labor Force Participation Decisions over the Business Cycle." Unpublished manuscript, 2006.

Keane, M., R. Moffitt, and D. Runkle. "Real Wages over the Business Cycle: Estimating the Impact of Heterogeneity with Micro Data." Journal of Political Economy, 96, 1988, 1232-66.

Kimmel, J., and K. Conway. "Who Moonlights and Why? Evidence from the SIPP." Industrial Relations, 40, 2001, 89-120.

Kimmel, J., and L. M. Powell. "Moonlighting Trends and Related Policy Issues in Canada and the United States." Canadian Public Policy, 25, 1999, 207-31.

Mincer, J. "Labor Force Participation and Unemployment: A Review of Recent Evidence, " in Prosperity and Unemployment, edited by R. A. Gordon and M. S. Gordon. New York: John Wiley and Sons, 1966, 73-112.

Mishel, L., J. Bernstein, and J. Schmitt. The State of Working America 1998 99. Ithaca, NY: Cornell University Press, 1999.

Moulton, B. R. "Random Group Effects and the Precision of Regression Estimates." Journal of Econometrics, 32, 1986, 385-97.

Park, S., and D. Shin. "Explaining Pro-Cyclical Male-Female Wage Gaps." Economics Letters, 88, 2005, 231-35.

Partridge, M. "Moonlighting in a High Growth Economy: Evidence from U.S. State-Level Data." Growth and Change, 33, 2002, 424-52.

Paxson, C.H., and N. Sicherman. "The Dynamics of Dual Job Holding and Job Mobility." JOLE, 14, 1996, 35-93.

Renna, F. "Overtime and Moonlighting: A Cross-Country Analysis." Journal of Labor Research, 27, 2006, 575-91.

Renna, F., and R. L. Oaxaca. "The Economics of Dual Job Holding: A Job Portfolio Model of Labor Supply." IZA Discussion Paper No. 1915, 2006.

Spletzer, J. R. "Reexamining the Added Worker Effect." Economic Inquiry, 35, 1997, 417-27.

Stinson, J. F. Jr. "Moonlighting: A Key to Differences in Measuring Employment Growth." Monthly Labor Review, 110(2), 1987, 30-31.

--. "Multiple Job Holding Up Sharply in the Eighties." Monthly Labor Review, 113(7), 1990, 3-10.

Tremblay, C. H. "Wage Patterns of Women over the Business Cycle." Quarterly Review of Economics and Business, 30, 1990, 90-101.

Vroman, W., and D. S. Nightingale. "Multiple Job Holding and Unemployment Insurance." Unpublished report, Urban Institute, 1996.

Wooldridge, J. M. Econometric Analysis of Cross Section and Panel Data, 2nd ed. Cambridge, MA: MIT, 2008.

(1.) See, for example, The State of Working America, 2002-2003.

(2.) These data were compiled from the Current Population Survey by the Bureau of Labor Statistics.

(3.) These figures may be somewhat overstated as some job changers may be mistakenly categorized as moonlighters (p. 359).

(4.) Conway and Kimmel (1998) argue that if labor supply elasticity estimates were adjusted to reflect this additional labor supply adjustment, labor supply elasticities would be increased, although the magnitude of this increase would be small. In other words, labor supply is more elastic than current empirical estimates suggest.

(5.) For explanation of these two effects and recent evidence, see Mincer (1966) and Spletzer (1997), respectively.

(6.) Examples of research on this topic include Devereux (2001), Hart (2006), and Keane, Moffitt, and Runkle (1988).

(7.) The authors find that workers in the lowest earnings decile assign 3% of their earnings to UI payroll taxes, whereas for their counterparts in the highest earnings decile, only 0.5% of their earnings go to paying for such taxes.

(8.) On the topic of UI and moonlighting, see Vroman and Nightingale (1996).

(9.) Heineck and Schwarze (2004) compare secondary-job holding in Germany and the United Kingdom. This cross-country comparison allows them to draw some conclusions regarding the role that institutions might play in moonlighting outcomes. They conclude that while institutions matter, they are not a substantial factor in explaining moonlighting rates.

(10.) Renna and Oaxaca (2006) focus on moonlighting workers who report no primary-job hours constraints and whose multiple-job holding choice reflects a "personal preference for job differentiation" (p. 1). Their job portfolio model confirms Conway and Kimmel's (1998) finding of a stronger wage elasticity of labor supply on the second job.

(11.) British moonlighting differs from U.S. moonlighting in three important ways: Brits have a higher moonlighting rate, they display greater moonlighting persistence over time, and their average secondary-job wages are much higher than their average primary-job wages.

(12.) Hotchkiss and Robertson (2006) show that female labor force participation exhibits a much stronger procyclicality than that of males, which they explain is due to the females' higher reservation wage. In fact, controlling for education, men are half as responsive to labor market conditions so that, even if they become unemployed during an economic downturn, they are half as likely to leave the labor force.

(13.) The NLSY Geo-coded data file and documentation are available at http://www.bls.gov/nls/nlsgeo97.htm. The geo-coded file was obtained under contract agreement no. 03-77.

(14.) Park and Shin (2005) and Tremblay (1990) use these data to examine the cyclicality of real wages by gender.

(15.) Robust standard errors are computed to correct for the heteroskedasticity that may affect our estimates.

(16.) We deflate hourly wages using the consumer price index for all urban consumers, not seasonally adjusted, with base period 1982 1984, which was retrieved from hnp://www.bls.gov/cpi/home.htm.

(17.) In this manner, we avoid counting as moonlighting job transitions during which the former and new jobs overlap briefly.

(18.) We use 5-yr averages. For the year 1980, we use data from 1979 to 1983 (the survey started in 1979; hence, data for 1978 are unavailable). For 1990, we average data from 1988 to 1992, and for the year 2000, averages from the 1998-2002 period are computed. In this last period, we include 3 yr instead of 5 yr because the NLSY79 survey switched from an annual to a biennial survey at that time.

(19.) Conway and Kimmel (1998) provide a detailed derivation of this theoretical framework.

(20.) A second potential disadvantage of the Tobit and two-part selection models is their reliance on normality and homoskedasticity in the latent variables. However, as noted by Wooldridge (2008), neither conditional normality nor heteroskedasticity affect the unbiasedness or consistency of the OLS estimates, and as a result, for reasonable deviations from these assumptions, the Tobit model still provides good estimates.

(21.) It should be noted that a fixed-effects Tobit model is not estimated as we lack a sufficient statistic to condition the fixed effects out of the likelihood function.

(22.) We opted not to include year dummies because, to the extent that our data come from a single cohort of individuals, the population is not likely to have a different distribution over time, and as such, the year dummies would likely be picking up much of the cyclical variation that is the focus of our research.

(23.) These estimates (as well as the ones corresponding to the number of hours moonlighted) are computed adding up the marginal effects corresponding to the time period dummy and its interaction term evaluated at the mean growth rate of nonfarm employment in Appendix Table A 1.

(24.) Kimmel and Conway (2001) compare primary-and secondary-job wages across age and education categories, primary- and secondary-job occupations, as well as family income status and consistently find that on average, primary-job wages exceed secondary-job wages. Relevant to this question is the recent work by Hart (2006), who examines the real-wage cyclicality of full- versus part-time workers.

CATALINA AMUEDO-DORANTES and JEAN KIMMEL *

* We are grateful for comments received at the Society for Labor Economists meetings as well as the feedback from anonymous referees.

Amuedo-Dorantes: Professor, Department of Economics, San Diego State University, San Diego, CA 92182. Phone 1-(619)-594-1663, Fax 1-(619)-594-5062, E-mail camuedod@mail.sdsu.edu; and IZA, Bonn, Germany.

Kimmel: Associate Professor, Department of Economics, Western Michigan University, Kalamazoo, MI 49008. Phone (269) 387-5541, Fax (269) 387-5637, E-mail jean.kimmel@wmich.edu; and IZA, Bonn, Germany.

TABLE 1
Working and Moonlighting Rates by Gender

 1980 1990

Variables Working Moonlighting Working Moonlighting

Men 0.72 0.05 0.81 0.08
Women 0.79 0.05 0.91 0.09

 2000

Variables Working Moonlighting

Men 0.82 0.07
Women 0.91 0.07

Notes: Authors' tabulations using the NLSY79.

TABLE 2
Single- versus Dual-Job Holders (Only Includes Workers)

 1980

Variables One Job Moonlighting

(A) Male
 Percent white 0.73 0.81
 Percent black 0.21 0.15
 Highest grade completed 11.86 12.82
 Married 0.21 0.14
 No. of children 0.24 0.09
 Urban 0.79 0.82

(B) Female
 Percent white 0.71 0.81
 Percent black 0.24 0.14
 Highest grade completed 11.49 12.05
 Married 0.12 0.11
 No. of children 0.10 0.10
 Urban 0.78 0.82

 1990

Variables One Job Moonlighting

(A) Male
 Percent white 0.69 0.71
 Percent black 0.24 0.25
 Highest grade completed 13.02 13.64
 Married 0.56 0.46
 No. of children 1.22 0.99
 Urban 0.79 0.81

(B) Female
 Percent white 0.69 0.70
 Percent black 0.25 0.25
 Highest grade completed 12.66 13.12
 Married 0.50 0.49
 No. of children 0.79 0.85
 Urban 0.79 0.80

 2000

Variables One Job Moonlighting

(A) Male
 Percent white 0.63 0.64
 Percent black 0.30 0.30
 Highest grade completed 13.31 13.73
 Married 0.58 0.54
 No. of children 1.55 1.67
 Urban 0.76 0.77

(B) Female
 Percent white 0.67 0.64
 Percent black 0.27 0.32
 Highest grade completed 13.07 13.27
 Married 0.61 0.57
 No. of children 1.27 1.18
 Urban 0.74 0.76

Notes: Authors' tabulations using the NLSY79.

TABLE 3
Random-Effects Tobit Model of Moonlighting

 Standard
Variables Coefficient Error

(A) Male

 Personal and family characteristics
 Black 2.355 *** 0.927
 Other race 1.828 1.776
 Age 0.083 0.154
 Highest grade 1.307 *** 0.177
 Married -3.012 *** 0.784
 Young children -4.602 *** 0.973
 No. of children 0.753 * 0.428
 Health limitations 0.264 1.511
 Past nonlabor income -3.05E-06 8.57E-06

 Primary-job characteristics
 Real hourly wage -0.005 0.025
 Public sector job 1.107 0.929
 Professional, technical, clerical -2.132 *** 0.787
 Craftsmen, operatives, laborers -1.838 1.200
 Tenure 0.278 *** 0.012
 Tenure squared -5.21E-05 *** 2.46E-06

 Regional and time-related factors
 Urban 1.179 0.940
 Northeast -29.059 * 15.682
 South -12.886 11.897
 West 12.973 17.457
 Time period 1979-1985 10.968 *** 3.788
 Time period 1986-1992 8.720 *** 2.506
 Time period 1993-1999 8.520 *** 2.528
 Growth nonfarm employment 0.476 0.699
 Period 1979-1985 x Growth NFE -0.780 0.740
 Period 1986-1992 x Growth NFE -0.366 0.757
 Period 1993-1999 x Growth NFE -1.584 0.981
 State per capita income -0.001 *** 3.02E-04

 Regression fit statistics
 Observations 11,391
 Groups 3,907
 Wald [chi square](78) 944.22
 Log likelihood -32,470.957

(B) Female

 Personal and family characteristics
 Black 3.419 *** 1.043
 Other race -1.518 1.875
 Age -0.040 0.177
 Highest grade 1.123 *** 0.192
 Married -0.777 1.018
 Young children -0.470 1.236
 No. of children 0.495 0.540
 Health limitations 0.125 2.150
 Past nonlabor income 5.22E-06 1.04E-05

 Primary-job characteristics
 Real hourly wage -0.054 ** 0.027
 Public sector job 5.809 *** 1.159
 Professional, technical, clerical -2.707 *** 1.072
 Craftsmen, operatives, laborers -6.116 *** 0.992
 Tenure 0.325 *** 0.013
 Tenure squared -6.01E-05 *** 2.48E-06

 Regional and time-related factors
 Urban -0.495 1.041
 Northeast -14.457 14.246
 South 12.848 14.969
 West 15.476 20.581
 Time period 1979-1985 25.013 *** 4.388
 Time period 1986-1992 19.311 *** 2.950
 Time period 1993-1999 9.738 *** 2.944
 Growth nonfarm employment 1.435 * 0.820
 Period 1979-1985 x Growth NFE -2.090 ** 0.865
 Period 1986-1992 x Growth NFE -2.184 *** 0.884
 Period 1993-1999 x Growth NFE -1.820 * 1.101
 State per capita income -0.002 *** 3.50E-04

 Regression fit statistics
 Observations 13.7448
 Groups 4,424
 Wald [chi square](78) 1,114.97
 Log likelihood -36,825.782

 ME on Prob ME on
Variables (Y > 0) E(Y|Y > 0)

(A) Male

 Personal and family characteristics
 Black 0.030 0.867
 Other race 0.023 0.675
 Age 0.001 0.030
 Highest grade 0.017 0.475
 Married -0.038 -1.088
 Young children -0.058 -1.635
 No. of children 0.010 0.274
 Health limitations 0.003 0.096
 Past nonlabor income -3.88E-08 -1.11E-06

 Primary-job characteristics
 Real hourly wage -6.00E-OS -0.002
 Public sector job 0.014 0.405
 Professional, technical, clerical -0.027 -0.777
 Craftsmen, operatives, laborers -0.023 -0.659
 Tenure 0.004 0.101
 Tenure squared -6.60E-07 -1.89E-05

 Regional and time-related factors
 Urban 0.015 0.429
 Northeast -0.342 -8.949
 South -0.162 -4.556
 West 0.162 5.095
 Time period 1979-1985 0.138 4.142
 Time period 1986-1992 0.110 3.232
 Time period 1993-1999 0.107 3.264
 Growth nonfarm employment 0.006 0.173
 Period 1979-1985 x Growth NFE -0.010 -0.284
 Period 1986-1992 x Growth NFE -0.005 -0.133
 Period 1993-1999 x Growth NFE -0.020 -0.576
 State per capita income -1.34E-05 -3.85E-04

 Regression fit statistics
 Observations
 Groups
 Wald [chi square](78)
 Log likelihood

(B) Female

 Personal and family characteristics
 Black 0.035 1.167
 Other race -0.015 -0.505
 Age -4.10E-04 -0.013
 Highest grade 0.011 0.378
 Married -0.008 -0.261
 Young children -0.005 -0.158
 No. of children 0.005 0.166
 Health limitations 0.001 0.042
 Past nonlabor income 5.33E-08 1.76E-06

 Primary-job characteristics
 Real hourly wage -0.001 -0.018
 Public sector job 0.060 2.022
 Professional, technical, clerical -0.028 -0.903
 Craftsmen, operatives, laborers -0.062 -2.050
 Tenure 0.003 0.109
 Tenure squared -6.14E-07 -2.02E-05

 Regional and time-related factors
 Urban -0.005 -0.166
 Northeast -0.144 -4.529
 South 0.131 4.446
 West 0.158 5.592
 Time period 1979-1985 0.253 9.027
 Time period 1986-1992 0.196 6.733
 Time period 1993-1999 0.100 3.440
 Growth nonfarm employment 0.015 0.483
 Period 1979-1985 x Growth NFE -0.021 -0.703
 Period 1986-1992 x Growth NFE -0.022 -0.734
 Period 1993-1999 x Growth NFE -0.019 -0.612
 State per capita income -1.85E-05 -0.001

 Regression fit statistics
 Observations
 Groups
 Wald [chi square](78)
 Log likelihood

Notes: All regressions include a constant term, state dummies, and a
time trend. Sales and services are used as reference categories for
the primary-job occupation. ME, marginal effects; NFE, nonfarm
employment. ***, **, and * signify statistically different from zero
at the l%, 5%, and 10% levels, respectively.

TABLE 4
Tobit Estimates of Moonlighting Cyclicality
among Working Men and Women

 Men

Group Computation p h (min)

Cyclicality in [meg.sub.grnfe] + -0.004 -0.111 (-7)
1979-1985 [me.sub.grnfe x
 period1979-1985]

Cyclicality in [meg.sub.grnfe] + 0.001 0.04 (2)
1986-1992 [me.sub.grnfe x
 period1986-1992]

Cyclicality in [meg.sub.grnfe] + -0.14 -0.403 (-24)
1993-1999 [me.sub.grnfe x
 period1993-1999]

Cyclicality in [meg.sub.grnfe] 0.006 0.173 (10)
2000-2002

 Women

Group p h (min)

Cyclicality in -0.004 ** -0.22 (13)
1979-1985

Cyclicality in -0.007 ** -0.251 (15)
1986-1992

Cyclicality in -0.006 -0.129 (8)
1993-1999

Cyclicality in 0.015 * 0.0483 * (29)
2000-2002

Notes: For the first three time periods in the table,
the significance levels correspond to those from a
joint significance [chi square](2) test.

** and * signify statistically different from zero
at the 5% and 10% levels, respectively.